Implementation of Distance Based outlier detection (DOD) baseline and dynamic K and R alternative (Dyn).

Dyn method calculates best k and R parameter in each window, based on objective function:

conda env create --file environment.yaml
conda activate ADBOD

X_data, label=data_and_labels(sequence_len=10,dataset="./data/YAHOO/Yahoo_A1real_53_data.out")


# Application of generating score in online fashion for all data, applyig sliding window:
dod_clf = DOD(k=50, R=10, window=200, slide=100)
dyn_clf = Dyn(window=200, slide=100)

DOD_anomalies=dod_clf.fit(X_data)
Dyn_anomalies=dyn_clf.fit(X_data)

plot(DOD_anomalies, Dyn_anomalies, label)

Output:

main.py: Example of usage of the two algorithms in online fashion.

ExperimentTSB.py: a helper class to conduct experiments using the two algorithms. RunExperiments.py: Tesing multiple hyper-parameters for DOD and Dyn and store the results under resulst Folder (results_final.csv file for Dyn and resultsKR_final.csv for DOD). /results/readResults.py: Used to search results from DOD and Dyn experiment, and to produce plot with summarized results.

data/ : data used for the evaluation Techniques/ : Implementation of Dyn and DOD

We use Window functionality to transform uni-variate to multivariate data using sub-sequences, using TSB implementation under the foldt TSB_UAD_code/.

@INPROCEEDINGS{10555107,
  author={Giannoulidis, Apostolos and Nikolaidis, Nikodimos and Gounaris, Anastasios},
  booktitle={2024 IEEE 40th International Conference on Data Engineering Workshops (ICDEW)}, 
  title={Parameter-free Streaming Distance-based Outlier Detection}, 
  year={2024},
  volume={},
  number={},
  pages={102-106},
  keywords={Sensitivity;Conferences;Data engineering;Anomaly detection;Streams;distance-based outlier detection;data streams},
  doi={10.1109/ICDEW61823.2024.00019}}